Zhe_2014JointSummits_v6

Categorizing the Rela/onships between Structurally Congruent Concepts from Pairs of Terminologies for Seman/c Harmoniza/on

Zhe He, PhD1,2 James Geller, PhD2, Gai Elhanan, MD3

1Department of Biomedical Informa3cs, Columbia University 2Department of Computer Science, New Jersey Ins3tute of Technology

3Halfpenny Technologies, Inc

2014 AMIA Summit on Transla5onal Bioinforma5cs April 8, 2014

Disclosure •  Zhe He discloses that he has no rela/onships with commercial

interests. •  James Geller discloses that he has no rela/onships with

commercial interests. •  Gai Elhanan discloses that he has no rela/onships with

commercial interests.

2

Learning Objec/ve

•  Use structural method to find poten/al concepts for enriching the conceptual content of a biomedical terminology

3

Overview •  Mo/va/on –  Exploring structural method for seman/c harmoniza/on

•  Background –  Importance of the conceptual content of SNOMED CT

•  Methods –  Structural matching of pairs of terminologies in the UMLS

•  Results –  Reusable knowledge can be derived by structural matching, including discovery of possible synonym

•  Limita/ons and Future Work •  Conclusions

4 4

Mo/va/on

� Need of well-‐developed and well-‐maintained terminologies

� NLP tools that process clinical text need a terminology with fruiXul concepts and synonyms.

�  Complex clinical research texts require combined use of mul/ple terminologies (Weng et al. 2010)

�  Terminologies need harmoniza/on to achieve seman/c interoperability (Bi]ner et al. 2005)

5 5

6

Seman/c Harmoniza/on Between Different Terminologies

•  The BRIDG model as a user-‐centered harmoniza/on framework (Weng et al. 2010)

•  Harmonized exis/ng /me ontologies for annota/ng temporal rela/on in clinical narra/ves (Tao et al. 2011)

•  Seman/c harmoniza/on efforts have recently been extended for various terminologies •  SNOMED CT and LOINC (AMIA 2013 Informa/cs Year in Review) •  SNOMED CT and ICD 11 (Rodrigues et al. 2013)

7

Importance of Conceptual Content of SNOMED CT

•  SNOMED CT is going to be a major terminology for EHR encoding of diagnoses and problem lists by 2015

•  SNOMED CT has many problems! •  Top two men/oned deficiencies of SNOMED CT (Elhanan 2011):

– Missing concepts – 23% – Missing synonyms – 17%

•  Users will expect SNOMED CT to have correct synonyms and sufficient concepts to be used in EHR

8

Leveraging Common Structure of Pairs of Terminologies in the UMLS

•  UMLS (Unified Medical Language System): –  More than 170 source terminologies, 8.9 million terms, 2012AB release

Breast Cancer (NCIt)

Carcinoma of breast (SNOMED)

Carcinoma breast (MedDRA)

Breast Carcinoma (UMLS)

etc.

Structurally Congruent Concepts

1)  X and Y are alterna/ve classifica/ons 2)  X can be a parent of Y 3)  Y can be a parent of X 4)  X and Y are synonymous 5)  Structural errors in Terminology 1 6)  Structural errors in Terminology 2

9

Concept A

Concept X

Concept B Concept B

Concept Y same concept

same concept

Concept A

X does not occur in Terminology 2

Y does not occur in Terminology 1

Cycles were eliminated during the processing

IS_A

IS_A

IS_A

IS_A

Alterna/ve Classifica/on

Concept A

Concept X Concept Y

same concept

same concept

Concept A

Classified by assump3on of Terminology 2 Designer

Concept B Concept B

Classified by assump3on of Terminology 1 Designer

10 10

IS_A

IS_A

IS_A

IS_A

IS_A

IS_A

X can be a parent of Y

11

Concept X

Concept Y

Y can be a parent of X Concept Y

Concept X

Parent-‐Child Rela/onship

11

IS_A

IS_A

12

X and Y are synonymous but have not been iden/fied by the UMLS

Concept X (Synonym: Y)

Synonymous

12

META Terminologies in this Study

•  Metathesaurus terminologies with “PAR” rela/onship and “inverse_isa” rela/onship a]ribute were chosen

•  SCTUSX and UWDA were excluded •  Reference Terminologies (Terminology 1): – MEDCIN (MEDCIN) –  Na/onal Cancer Ins/tute Thesaurus (NCIt) –  Gene Ontology (GO) – Medical En//es Dic/onary (CPM) –  UMDNS: product category thesaurus (UMD) –  Founda/onal Model of Anatomy Ontology (FMA)

•  Terminology 2: SNOMED CT

13

14

Evalua/on: Pairs of Congruent Concepts Found

Reference Terminology

Size of Terminology # of Pairs of Congruent Concepts

Sample Size

MEDCIN 279529 655 70 NCI 95523 582 70 FMA 82062 116 70 UMD 15956 18 18 GO 61925 6 6 CPM 3078 7 7 Total -‐-‐ 1384 241

15

Review Results for Pairs of Congruent Concepts


Sample Size Alterna5ve Classifica5on

Y à X X à Y

MEDCIN 70 44 10 7 NCI 70 38 12 6 GO 6 2 -‐-‐ 4 CPM 7 5 -‐-‐ -‐-‐ UMD 18 9 1 -‐-‐ FMA 70 45 13 4 Total 241 143 36 21 Percentage 100% 59.3% 14.9% 8.7%

23.6%

Co-‐author GE reviewed the sample

16

Review Results for Pairs of Congruent Concepts


Sample Size

Error in Termi-‐ nology 1

Error in Termi-‐ nology 2

Synonym

MEDCIN 70 -‐-‐ 1 8 NCI 70 -‐-‐ 3 11 GO 6 -‐-‐ -‐-‐ -‐-‐ CPM 7 -‐-‐ -‐-‐ 2 UMD 18 -‐-‐ -‐-‐ 8 FMA 70 2 -‐-‐ 6 Total 181 2 4 35 Percentage 100% 0.8% 1.7% 14.5%

17

Example: Alterna/ve Classifica/on

17

Structure of posterior intercostal vein, C0226639

Eleventh right posterior intercostal vein, C0501203


Eleventh posterior intercostal vein, C0506471

Lower right posterior intercostal veins, C1283497


FMA3_1 SNOMEDCT_2012_07_31

17

Making Explicit an Implicit Assump/on of the Two Original Terminology Designers



Eleventh posterior intercostal vein, C0506471

Posterior intercostal vein classified by position

SNOMEDCT_2012_07_31


Posterior intercostal vein classified by ordinality


Lower right posterior intercostal veins, C1283497

FMA3_1

18 18

19

Example: Parent of the Other

Sign and Symptoms, C0037088

Finding by Site or System, C1333618 Finding by site, C1290906

Integumentary System Finding, C1291044

FMA3_1 SNOMEDCT_2012_07_31



Sugges/on: Concept Import

Finding by Site or System, C1333618

Finding by site, C1290906



20 20

Example: Synonym of the Other

21

Chemicals, C0220806

Organic Chemicals, C0029224

Chemicals, C0220806

Chemical Viewed Structurally, C1254350

Chemical categorized structurally, C0729761


CPM2003 SNOMEDCT_2012_07_31

21

Sugges/on: Merge Two Synonymous Concepts

Chemicals, C0220806

Chemical categorized structurally, C0729761 Synonym: Chemical Viewed Structurally


22

Limita/ons

23

�  Harmoniza/on cannot be done without the consent of terminology curators

�  All the terminologies are in UMLS Rich Release format

23

Future Work

24

� More complex configura/ons: more intermediate concepts

�  Algorithm to iden/fy the rela/onships between intermediate concepts in complex configura/ons

�  Pairs of any two terminologies

24

Conclusions

25

�  Six reference terminologies of the UMLS vs. SNOMED

�  In a sample of 241 congruency pairs �  143 out of 241 (59.3%) concept pairs: alterna/ve classifica/on. �  47 out of 241 (23.6%) concept pairs: parent-‐child rela/onships. �  35 (14.5%) new synonyms �  Three pairs of concepts indicated errors

�  Take home message:

�  A semi-‐automated way based on common structure of the UMLS may complement exis/ng human-‐expert centered methods to find poten/al concepts for import and export to a terminology.

25

26

Acknowledgment

We would like to thank: Dr. Yehoshua Perl and Dr. Chunhua Weng for sharing their insights and giving feedback for this work

References (1)

27

Weng C, Tu SW, Sim I, Richesson R. Formal representa/on of eligibility criteria: a literature review. J Biomed Inform. 2010;43(3):451-‐67. Bi]ner T, Donnelly M, S. W. Ontology and seman/c interoperability. In: Prosperi D, Zlatanova S, editors. Large-‐scale 3D data integra/on: Problems and challenges: CRCpress (Tailor & Francis); 2005: 139-‐60. Tao C, Solbrig HR, Chute CG. CNTRO 2.0: A Harmonized Seman/c Web Ontology for Temporal Rela/on Inferencing in Clinical Narra/ves. AMIA Summits Transl Sci Proc. 2011;2011:64-‐8. Rodrigues JM, Schulz S, Rector A, Spackman KA, Üstün B, Chute CG, Della Mea V, Millar J, Persson KB. Sharing Ontology between ICD 11 and SNOMED CT will enable Seamless Re-‐use and Seman/c Interoperability. Stud Health Technol Inform. 2013;192:343-‐6 Elhanan G, Perl Y, Geller J. A survey of SNOMED CT direct users, 2010: impressions and preferences regarding content and quality. J Am Med Inform Assoc. 2011; Suppl 1:i33-‐44

27

References (2)

28

Kumar A, Smith B, Novotny DD. Biomedical informa/cs and granularity. Comp Funct Genomics. 2004;5(6-‐7):501-‐8. Weng C, Gennari JH, Fridsma DB, User-‐centered Seman/c Harmoniza/on: A Case Study, J Biomed Inform, 2007, 40(3):353-‐64 Schulz S, Boeker M, Stenzhorn H. How Granularity Issues Concern Biomedical Ontology Integra/on. In Proceedings of the Interna/onal Congress of the European Federa/on for Medical Informa/cs (MIE 2008). Gothenburg, Sweden; 2008: 863-‐68. Rector A, Rogers J, Bi]ner T. Granularity, scale and collec/vity: when size does and does not ma]er. J Biomed Inform. 2006 Jun;39(3):333-‐49. Mougin F, Bodenreider O. Approaches to elimina/ng cycles in the UMLS Metathesaurus: naive vs. formal. AMIA Annu Symp Proc. 2005:550-‐4.

28

Thanks!

Zhe He, PhD [email protected]

Zhe_2014JointSummits_v6

Documents

Transcript of Zhe_2014JointSummits_v6